There are demands for realizing a physical support robot for use in an ordinary living environment such as in a house. In order to realize such a robot, it is necessary to realize processes such as detecting the position and the posture of an object, recognizing the object, detecting the motion of the object, and recognizing the environment.
For performing such processes, methods using cameras have been widely used in conventional methods. This is realized by performing an image process, such as pattern matching, optical flow detection and feature point extraction, on images captured by a camera.
It is known that in an ordinary living environment such as in a house, the accuracy and the detection rate of these image processes are deteriorated by the influence of specular reflection/shadows due to light source variations. In view of this, researches have been conducted on image processes that are robust against the influence of light source variations (e.g., photometric linearization), using a reflection model such as a dichromatic reflection model or a Lambertian model (for example, Non-Patent Documents 1 and 2).
In Non-Patent Document 1, the color of an object observed under an outdoor environment is considered as being composed of an all-sky light component that does not vary over time, and a direct sunlight component whose brightness varies rapidly when the direct sunlight is blocked by a cloud. When performing a background subtraction method for detecting a moving object, the mixing ratio of the direct sunlight component, which varies substantially over time, is estimated, thereby realizing a detection of a moving object that is not influenced by shadows and the luminance variation of the light source.
Non-Patent Document 2 uses a plurality of images through which the light source position varies to produce linearized images each being a synthetic image that is composed of only diffuse reflection and does not include specular reflection or shadows. By using such linearized images, it is possible to realize an image process that is not influenced by specular reflection.
Non-Patent Document 1: Akihiro Okumura, Yoshio Iwai, Masahiko Yachida, “Moving Object Detection in the Open Air Environment”, Meeting on Image Recognition and Understanding (MIRU 2000), vol. II, pp. 307-312, 2000
Non-Patent Document 2: Yasunori Ishii, Koutaro Fukui, Yasuhiro Mukaigawa, Takeshi Shakunaga, “Photometric Linearization Based on Classification of Photometric Factors”, Journal of Information Processing Society of Japan, vol. 44, no. SIG5 (CVIM6), pp. 11-21, 2003